This continuation proposal requests four years of additional support to complete an empirical analysis of data on divorce and divorce settlements from the NLS'72. The current empirical project builds on a theoretical model of marriage and divorce which served as the conceptual basis for the design of a questionnaire module for the 5th Follow-Up of the NLS'72 in a related NICHD project. In the current project a lengthy study of divorce settlements has been completed using these data. The primary aim of this continuation proposal is to examine the determinants of the probability of divorce, including the influence of factors leading to early marriage and cohabitation, the influence of market and marriage-specific capital and childbearing during the marriage, and the role of expectations concerning the divorce settlement and opportunities outside the marriage for each spouse. A preliminary analysis of the probability of divorce is contained in Weiss/Willis(1989), but main questions concerning the determinants of divorce that will be pursued in this project: (1) What role does unexpected variation in spouse characteristics (i.e., "surprise") play in causing divorce? (2) To what extent do children inhibit fertility? (3) To what extent does investment in marketable human capital by wives increase the risk of divorce or, conversely, to what extent is such investment motivated by the prospect of divorce? (4) Does the expected divorce settlement (and the laws governing divorce settlements) influence the probability of divorce and why? (5) Do early marriage and cohabitation have a causal impact on subsequent marital instability? These questions raise difficult methodological issues concerning endogeneity and the measurement of expectations which must be solved in order to sort out the causal relationship underlying the correlates of divorce. Our empirical strategy begins with the estimation of reduced form hazard models of divorce in which a variety of exogeneity tests will be conducted. A structural model of divorce in which a couple's decision to divorce and other endogenous choices in each period incorporate their expectation about the course of future endogenous variables, based on their current information, will be estimated using techniques recently developed by Hotz/Miller.